1. An optimized sampling procedure was used to collect georeferenced vegetation plot data. A multivariate analysis was then used to define vegetation types that could be related to different colonization stages and environmental contexts.
2. Color infrared aerial photographs were then used to produce a baseline vegetation map. This map was then integrated into a data base along with other environment factors known to control plant colonization processes, such as climate (wind, temperature), physical landscape components (habitat characteristics) and morphodynamic processes (runoff).
3. A Bayesian model using conditional probabilities was used to identify the primary environmental habitats corresponding to the different vegetation types.
This protocol was tested on the fore field of the Midre Lovénbreen (Svalbard) glacier where several vegetation belts correspond to well defined stages of deglaciation and corresponding local conditions such as microtopography, microclimate and runoff dynamics.